Su Wang

State University of New York, New York, New York, United States

Are you Su Wang?

Claim your profile

Publications (32)26.5 Total impact

  • [Show abstract] [Hide abstract]
    ABSTRACT: Bladder cancer is the fifth cause of cancer deaths in the United States. Virtual cystoscopy (VC) can be a screening means for early detection of the cancer using non-invasive imaging and computer graphics technologies. Previous researches have mainly focused on spiral CT (computed tomography), which invasively introduces air into bladder lumen for a contrast against bladder wall via a small catheter. However, the tissue contrast around bladder wall is still limited in CT-based VC. In addition, CT-based technique carries additional radiation. We have investigated a procedure to achieve the screening task by MRI (magnetic resonance imaging). It utilizes the unique features of MRI: (1) the urine has distinct T1 and T2 relaxation times as compared to its surrounding tissues, and (2) MRI has the potential to obtain good tissue contrast around bladder wall. The procedure is fully non-invasive and easy in implementation. In this paper, we proposed a MRI-based VC system for computer aided detection (CAD) of bladder tumors. The proposed VC system is an integration of partial volume-based segmentation containing texture information and fast marching-based CAD employing geometrical features for detecting of bladder tumors. The accuracy and efficiency of the integrated VC system are evaluated by testing the diagnoses against a database of patients.
    No preview · Article · Aug 2010 · Proceedings of SPIE - The International Society for Optical Engineering
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we propose a coupled level set (LS) framework for segmentation of bladder wall using T<sub>1</sub>-weighted magnetic resonance (MR) images with clinical applications to virtual cystoscopy (i.e., MR cystography). The framework uses two collaborative LS functions and a regional adaptive clustering algorithm to delineate the bladder wall for the wall thickness measurement on a voxel-by-voxel basis. It is significantly different from most of the pre-existing bladder segmentation work in four aspects. First of all, while most previous work only segments the inner border of the wall or at most manually segments the outer border, our framework extracts both the inner and outer borders automatically except that the initial seed point is given by manual selection. Secondly, it is adaptive to T<sub>1</sub>-weighted images with decreased intensities in urine, as opposed to enhanced intensities in T<sub>2</sub>-weighted scenario and computed tomography. Thirdly, by considering the image global intensity distribution and local intensity contrast, the defined image energy function in the framework is more immune to inhomogeneity effect, motion artifacts and image noise. Finally, the bladder wall thickness is measured by the length of integral path between the two borders which mimic the electric field line between two iso-potential surfaces. The framework was tested on six datasets with comparison to the well-known Chan-Vese (C-V) LS model. Five experts blindly scored the segmented inner and outer borders of the presented framework and the C-V model. The scores demonstrated statistically the improvement in detecting the inner and outer borders.
    Full-text · Article · Apr 2010 · IEEE Transactions on Medical Imaging
  • [Show abstract] [Hide abstract]
    ABSTRACT: With the development of computer-aided polyp detection towards virtual colonoscopy screening, the trade-off between detection sensitivity and specificity has gained increasing attention. An optimum detection, with least number of false positives and highest true positive rate, is desirable and involves interdisciplinary knowledge, such as feature extraction, feature selection as well as machine learning. Toward that goal, various geometrical and textural features, associated with each suspicious polyp candidate, have been individually extracted and stacked together as a feature vector. However, directly inputting these high-dimensional feature vectors into a learning machine, e.g., neural network, for polyp detection may introduce redundant information due to feature correlation and induce the curse of dimensionality. In this paper, we explored an indispensable building block of computer-aided polyp detection, i.e., principal component analysis (PCA)-weighted feature selection for neural network classifier of true and false positives. The major concepts proposed in this paper include (1) the use of PCA to reduce the feature correlation, (2) the scheme of adaptively weighting each principal component (PC) by the associated eigenvalue, and (3) the selection of feature combinations via the genetic algorithm. As such, the eigenvalue is also taken as part of the characterizing feature, and the necessary number of features can be exposed to mitigate the curse of dimensionality. Learned and tested by radial basis neural network, the proposed computer-aided polyp detection has achieved 95% sensitivity at a cost of average 2.99 false positives per polyp.
    No preview · Article · Mar 2010 · Proceedings of SPIE - The International Society for Optical Engineering

  • No preview · Article · Jan 2010
  • [Show abstract] [Hide abstract]
    ABSTRACT: PURPOSE Virtual cystoscopy is a potential non-invasive alternative of the traditional invasive cystoscopy for evaluation of the entire bladder. This study aims to develop a level set (LS)-based algorithm, which can automatically extract the bladder wall by segmenting the inner and outer borders from T1-weighted magnetic resonance (MR) images (where the urine signal is suppressed to minimize partial volume effect on the detection of abnormality on the inner border), toward MR virtual cystoscopy. METHOD AND MATERIALS Based on the well-known Chan-Vese (C-V) model, two collaborative LS functions were proposed and initialized with the guidance of the anatomic geometry in the T1-weighted MR image and evolved according to the specially designed image and geometry energy terms in the LS functions. The evolution process also utilized the regional statistics between the initialized two LS functions and converged when the two collaborative LS functions arrived at the inner and outer borders of the bladder wall. Taking the delineated inner and outer borders as two iso-potential 3-D surfaces, the thickness of the bladder wall was measured without any ambiguity by calculating the length of the electric field line between the two surfaces. The wall thickness distribution can be visualized by color map on the bladder geometry and quantified for computer-aided detection of abnormality. RESULTS The presented algorithm was tested on seven patient studies with comparison to the original C-V model. The segmented inner and outer borders of the two methods were blindly scored by five experts. The statistical analysis on the scores showed that the algorithm significantly outperformed (p<0.0001) the C-V model on the outer border segmentation and approached to statistically significant for the inner border detection (p=0.066). The measured thickness distribution on the bladder wall was mapped with different colors, and the area surrounding a lesion of size greater than 8 mm was extraordinarily highlighted on the 3-D rendering result. CONCLUSION The proposed collaborative LS approach can automatically extract the bladder wall via segmenting the inner and outer borders from T1-weighted MR images with statistically significant improvement over the previous method, and can further uniquely map the wall thickness. CLINICAL RELEVANCE/APPLICATION MR virtual cystoscopy would have a significant impact on bladder cancer screening and tumor recurrence evaluation.
    No preview · Conference Paper · Dec 2009
  • Source
    Su Wang · Hongyu Lu · Zhengrong Liang
    [Show abstract] [Hide abstract]
    ABSTRACT: Voxels near tissue borders in medical images contain useful clinical information, but are subject to severe partial volume (PV) effect, which is a major cause of imprecision in quantitative volumetric and texture analysis. When modeling each tissue type as a conditionally independent Gaussian distribution, the tissue mixture fractions in each voxel via the modeled unobservable random processes of the underlying tissue types can be estimated by maximum a posteriori expectation-maximization (MAP-EM) algorithm in an iterative manner. This paper presents, based on the assumption that PV effect could be fully described by a tissue mixture model, a theoretical solution to the MAP-EM segmentation algorithm, as opposed to our previous approximation which simplified the posteriori cost function as a quadratic term. It was found out that the theoretically-derived solution existed in a set of high-order non-linear equations. Despite of the induced computational complexity when seeking for optimum numerical solutions to non-linear equations, potential gains in robustness, consistency and quantitative precision were noticed. Results from both synthetic digital phantoms and real patient bladder magnetic resonance images were presented, demonstrating the accuracy and efficiency of the presented theoretical MAP-EM solution.
    Full-text · Article · Jun 2009 · International Journal of Imaging Systems and Technology
  • Zhengrong Liang · Su Wang
    [Show abstract] [Hide abstract]
    ABSTRACT: The author affiliations in the first footnote of the above-named work are corrected.
    No preview · Article · Apr 2009 · IEEE Transactions on Medical Imaging
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: As a promising second reader of computed tomographic colonography (CTC) screening, the computer-aided detection (CAD) of colonic polyps has earned fast growing research interest. In this paper, we present a CAD scheme to automatically detect colonic polyps in CTC images. First, a thick colon wall representation, ie, a volumetric mucosa (VM) with several voxels wide in general, was segmented from CTC images by a partial-volume image segmentation algorithm. Based on the VM, we employed a level set-based adaptive convolution method for calculating the first- and second-order spatial derivatives more accurately to start the geometric analysis. Furthermore, to emphasize the correspondence among different layers in the VM, we introduced a middle-layer enhanced integration along the image gradient direction inside the VM to improve the operation of extracting the geometric information, like the principal curvatures. Initial polyp candidates (IPCs) were then determined by thresholding the geometric measurements. Based on IPCs, several features were extracted for each IPC, and fed into a support vector machine to reduce false positives (FPs). The final detections were displayed in a commercial system to provide second opinions for radiologists. The CAD scheme was applied to 26 patient CTC studies with 32 confirmed polyps by both optical and virtual colonoscopies. Compared to our previous work, all the polyps can be detected successfully with less FPs. At the 100% by polyp sensitivity, the new method yielded 3.5 FPs/dataset.
    Full-text · Article · Mar 2009 · Cancer Management and Research
  • Source
    Zhengrong Liang · Su Wang
    [Show abstract] [Hide abstract]
    ABSTRACT: This work presents an iterative expectation-maximization (EM) approach to the maximum a posteriori (MAP) solution of segmenting tissue mixtures inside each image voxel. Each tissue type is assumed to follow a normal distribution across the field-of-view (FOV). Furthermore, all tissue types are assumed to be independent from each other. Under these assumptions, the summation of all tissue mixtures inside each voxel leads to the image density mean value at that voxel. The summation of all the tissue mixtures' unobservable random processes leads to the observed image density at that voxel, and the observed image density value also follows a normal distribution (image data are observed to follow a normal distribution in many applications). By modeling the underlying tissue distributions as a Markov random field across the FOV, the conditional expectation of the posteriori distribution of the tissue mixtures inside each voxel is determined, given the observed image data and the current-iteration estimation of the tissue mixtures. Estimation of the tissue mixtures at next iteration is computed by maximizing the conditional expectation. The iterative EM approach to a MAP solution is achieved by a finite number of iterations and reasonable initial estimate. This MAP-EM framework provides a theoretical solution to the partial volume effect, which has been a major cause of quantitative imprecision in medical image processing. Numerical analysis demonstrated its potential to estimate tissue mixtures accurately and efficiently.
    Full-text · Article · Mar 2009
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: As a non-invasive bladder tumor screening approach, magnetic resonance imaging (MRI)-based virtual cystoscopy (VCys) has received increasing attention for a better soft tissue contrast compared to computer tomography (CT)-based VCys. In this paper, some preliminary work on segmenting the inner boundary of bladder wall from both T1- and T2- weighted MR bladder images were presented. Via an iterative maximum a posteriori expectation-maximization (MAPEM) approach, the tissue mixture fractions inside each voxel were estimated. Considering the partial volume effect (PVE) that MR images suffer from, the advantages of such mixture-based segmentation approach are (1) statistics-based tissue mixture model that shapes each tissue type as a normal-distributed random variable, (2) closed-form mathematical MAP-EM iterative solution, and (3) capability and efficiency of the estimated tissue mixture fractions in reflecting PVE. Given the extracted inner bladder wall, manipulations could be further taken, for each individual voxel located on the inner bladder wall, to identify the outer bladder wall prior to the measurement of wall thickness. Not limited to geometrical analysis, the consideration of PVE in the study of early stage abnormality on the mucosa in the scope of VCys is believed to provide more textural information in distinguishing from neighboring artifacts about the surface deformations that is due to bladder tumors.
    Full-text · Article · Feb 2009 · Proceedings of SPIE - The International Society for Optical Engineering
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: With the development of computer-aided detection of polyps (CADpolyp), various features have been extracted to detect the initial polyp candidates (IPCs). In this paper, three approaches were utilized to reduce the number of false positives (FPs): the multiply linear regression (MLR) and two modified machine learning methods, i.e., neural network (NN) and support vector machine (SVM), based on their own characteristics and specific learning purposes. Compared to MLR, the two modified machine learning methods are much more sophisticated and well-adapted to the data provided. To achieve the optimal sensitivity and specificity, raw features were pre-processed by the principle component analysis (PCA) in the hope of removing the second-order statistical correlation prior to any learning actions. The gain by the use of PCA was evidenced by the collected 26 patient studies, which included 32 colonic polyps confirmed by both optical colonoscopy (OC) and virtual colonoscopy (VC). The learning and testing results showed that the two modified machine-learning methods can reduce the number of FPs by 48.9% (or 7.2 FPs per patient) and 45.3% (or 7.7 FPs per patient) respectively, at 100% detection sensitivity in comparison with that of traditional MLR method. Generally, more than necessary number of features were stacked as input vectors to machine learning algorithms, dimensionality reduction for a more compact feature combination, i.e., how to determine the remaining dimensionality via PCA linear transform was considered and discussed in this paper. In addition, we proposed a new PCA-scaled data pre-processing method to help reduce the FPs significantly. Finally, fROC (free-response receiver operating characteristic) curves corresponding to three FP-reduction approaches were acquired, and comparative analysis was conducted.
    Full-text · Article · Feb 2009 · Proceedings of SPIE - The International Society for Optical Engineering
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Electronic colon cleansing (ECC) is an emerging technique developed to segment the colon lumen from a patient's abdominal computed tomography colonography (CTC) images. However, the residue stool and fluid tagged by contrast materials as well as mixed tissue distribution with partial volume (PV) effect impose several challenges for ECC, resulting in incomplete and overcomplete cleansings. To address the PV effect, this work investigated an improved maximum a posteriori expectation-maximization (MAP-EM) image segmentation algorithm which simultaneously estimates tissue mixture percentages within each image voxel and statistical model parameters for the tissue distribution. Given the segmented tissue mixture information beyond the image voxel level, not only the PV effect has been satisfactorily addressed as a particular case of tissue mixture problem, but incomplete and overcomplete ECC causes could also be maximally avoided. For clinical application to CTC that involves several issues transferring from theoretical analysis to practical validation, an innovative initialization procedure and refined estimation strategy were proposed to build an ECC pipeline based on the MAP-EM segmentation. The pipeline was evaluated based on 52 patient CTC studies, downloaded from the website of the Virtual Colonoscopy Screening Resource Center, by two radiologists. A noticeable improvement over the authors' previous ECC pipeline was documented. Several typical cases were also presented to show visually the improved performance of the presented ECC pipeline.
    Full-text · Article · Jan 2009 · Medical Physics
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: It has been widely accepted that for brain MR images, both the image density inhomogeneity (slowly-varying intensity changes across the field of view) and partial-volume effect (PVE) (more than one tissue type present in a single voxel) considerably reduce the accuracy and reliability of quantitative analysis for various clinical purposes. This paper presents a unified expectation-maximization (EM) approach, where PVE and intensity inhomogeneity are combined together into a built-in-one statistical model in additive and multiplicative formats. It assumes that each tissue type follows a conditionally-independent normal distribution, based on which the summation of all tissue contributions multiplied or added by the bias term leads to mean density value at each voxel. Meanwhile, the summation of all the tissue mixtures, which is unobservable but could be estimated via EM framework (many-to-one mapping), multiplied or added by the bias term would lead to the observed image density at each voxel. In doing so, both the inhomogeneity and tissue mixtures are updated voxel-by-voxel until the convergence of a stable solution. Comprehensive tests on simulated brain MR images strongly demonstrated the feasibilities of additive/multiplicative bias models and the effectiveness of the unified EM approach. In addition, additive and multiplicative bias field models reflect advantages in terms of stability and robustness.
    Full-text · Conference Paper · Nov 2008
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Recent researches have shown promise in applying KL transform to 4D gated sinogram for pre-reconstruction temporal smoothing and quasi-4D inversion of attenuated Radon transform. To achieve quantitative 4D reconstruction, this work aims to compensate for the major degradation factors, including distance-dependent collimator resolution variation and object-specific photon scatter, simultaneously within the KL framework. To alleviate the influence of cardiac motion on reconstruction, heart motion was classified into several groups based on inter-frame similarities and each group underwent a corresponding KL transform. In the KL domain, non-stationary Poisson noise was stabilized by Anscombe transform and treated by adaptive Wiener filtration. Scatter contribution to the primary energy window was then estimated and removed based on photon detection energy spectrum and the triple-energy-window acquisition formula after noise treatment. The scatter-corrected data was further subject to a depth-dependent deconvolution, based on the distance frequency relationship, with measured detector response kernel in the KL domain. The deconvoluted sinograms were reconstructed by inverting the attenuated Radon transform for each KL component and the 4D SPECT images were obtained by a corresponding inverse KL transform for each group. The simultaneous compensation strategy in the KL domain was tested by computer simulations from digital phantoms of 128 cubic array and clinical data from a patient. The adaptive KL transform for different groups consisting of frames with similar activity dynamics showed noticeable improvement over our previous work of using a single KL transform for all frames. Improvement was also seen by the adaptive noise treatment of all the KL components over previous work of discarding the higher-order components. Further improvement by considering the scatter and resolution variation was demonstrated.
    Full-text · Conference Paper · Nov 2008
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Computed tomography colonography (CTC) or CT-based virtual colonoscopy (VC) is an emerging tool for detection of colonic polyps. Compared to the conventional fiber-optic colonoscopy, VC has demonstrated the potential to become a mass screening modality in terms of safety, cost, and patient compliance. However, current CTC delivers excessive X-ray radiation to the patient during data acquisition. The radiation is a major concern for screening application of CTC. In this work, we performed a simulation study to demonstrate a possible ultra low-dose CT technique for VC. The ultra low-dose abdominal CT images were simulated by adding noise to the sinograms of the patient CTC images acquired with normal dose scans at 100 mA s levels. The simulated noisy sinogram or projection data were first processed by a Karhunen-Loeve domain penalized weighted least-squares (KL-PWLS) restoration method and then reconstructed by a filtered backprojection algorithm for the ultra low-dose CT images. The patient-specific virtual colon lumen was constructed and navigated by a VC system after electronic colon cleansing of the orally-tagged residue stool and fluid. By the KL-PWLS noise reduction, the colon lumen can successfully be constructed and the colonic polyp can be detected in an ultra low-dose level below 50 mA s. Polyp detection can be found more easily by the KL-PWLS noise reduction compared to the results using the conventional noise filters, such as Hanning filter. These promising results indicate the feasibility of an ultra low-dose CTC pipeline for colon screening with less-stressful bowel preparation by fecal tagging with oral contrast.
    Full-text · Article · Nov 2008 · IEEE Transactions on Nuclear Science
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Computed tomography (CT) has been well established as a diagnostic tool through hardware optimization and sophisticated data calibration. For screening purposes, the associated x-ray exposure risk must be minimized. An effective way to minimize the risk is to deliver fewer x-rays to the subject or lower the mAs parameter in data acquisition. This will increase the data noise. This work aims to study the noise property of the calibrated or preprocessed sinogram data in Radon space as the mAs level decreases. An anthropomorphic torso phantom was scanned repeatedly by a commercial CT imager at five different mAs levels from 100 down to 17 (the lowest value provided by the scanner). The preprocessed sinogram datasets were extracted from the CT scanner to a laboratory computer for noise analysis. The repeated measurements at each mAs level were used to test the normality of the repeatedly measured samples for each data channel using the Shapiro-Wilk statistical test merit. We further studied the probability distribution of the repeated measures. Most importantly, we validated a theoretical relationship between the sample mean and variance at each channel. It is our intention that the statistical test and particularly the relationship between the first and second statistical moments will improve low-dose CT image reconstruction for screening applications.
    Full-text · Article · Jun 2008 · Physics in Medicine and Biology
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Purpose Purpose A volume-based mucosa-based polyp candidate determination scheme for automatic polyp detection in computed colonography is presented in this paper. Methods Different from the one-layer mucosa that is widely accepted by the existing computer-aided detection methods, a thick mucosa region of 3–5 voxels wide is extracted, which excludes the direct applications of the traditional geometrical features. A fast marching-based adaptive gradient/curvature and weighted integral curvature along normal directions (WICND) are developed for this purpose, and polyp candidates are optimally determined by computing and clustering these fast marching-based adaptive geometrical features. Results By testing on 52 patients datasets in which 26 patients were found with polyps of size 4–22 mm, both the locations and number of polyp candidates detected by WICND and previously developed linear integral curvature (LIC) were compared. Not only the number of false positives (FPs) was reduced from 706 to 132 on average, but also the detection sensitivity has been slightly improved. Conclusions WICND outperformed LIC mainly by significantly reducing the number of FPs, which promises to release our burden of machine learning in the feature space, especially for those polyps smaller than 5 mm.
    Full-text · Article · Jun 2008 · International Journal of Computer Assisted Radiology and Surgery
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Computed tomography (CT) has been well established as a diagnostic tool through hardware optimization and sophisticated data calibration. For screening purposes, the associated X-ray exposure risk must be minimized. An effective way to minimize the risk is to deliver fewer X-rays to the subject or lower the mAs parameter in data acquisition. This will increase the data noise. This work aims to study the noise property of the calibrated or preprocessed sinogram data in Radon space as the mAs level decreases. An anthropomorphic torso phantom was scanned repeatedly by a commercial CT imager at five different mAs levels from 100 down to 17 (the lowest value provided by the scanner). The preprocessed sinogram datasets were extracted from the CT scanner to a laboratory computer for noise analysis. The repeated measurements at each mAs level were used to test the normality of the repeatedly measured samples for each data channel using the Shapiro-Wilk statistical test merit. We further studied the probability distribution of the repeated measures. Most importantly, we validated a theoretical relationship between the sample mean and variance at each channel. It is our intention that the statistical test and particularly the relationship between the first and second statistical moments will improve low-dose CT image reconstruction for screening applications.
    Full-text · Article · Apr 2008 · Proceedings of SPIE - The International Society for Optical Engineering
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Virtual cystoscopy (VC) is a developing noninvasive, safe, and low-cost technique for bladder cancer screening. Multispectral (T1- and T2-weighted) magnetic resonance (MR) images provide a better tissue contrast between bladder wall and bladder lumen comparing with computed tomography (CT) images. The intrinsic T1 and T2 contrast of the urine against the bladder wall eliminates the invasive air insufflation procedure which is often used in CT-based VC. We propose a new partial volume (PV) segmentation scheme with inhomogeneity correction to segment multispectral MR images for tumor screening by virtual cystoscopy. The proposed PV segmentation algorithm automatically estimates the bias field and segments tissue mixtures inside each voxel of MR images, thus preserving texture information. Experimental results indicate that the present scheme is promising towards mass screening by virtual cystoscopy means.
    Full-text · Article · Mar 2008 · Proceedings of SPIE - The International Society for Optical Engineering
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Quantitative volumetric measurement and feature analysis for various clinical applications require image segmentation. Most important clinical features are derived from the borders of a region of interest, which reflects the shape characteristics and volumetric variation of the target. The partial volume (PV) effect renders a significant error for current hard segmentation which assigns a single tissue label to each image voxel. We have proposed an expectation-maximization (EM) framework for soft image segmentation which aims to quantify the tissue mixture percentages in each voxel. By imposing a priori Markov random field (MRF) penalty on the spatial distribution of each tissue mixture, the algorithm searches a maximum a posteriori (MAP) solution for the tissue model parameters of the given image and the tissue mixture percentages in each voxel. This work studied the sensitivity of the iterative MAP-EM algorithm to the initial estimate and the properties of its convergence for the estimation of the model parameters and tissue mixtures in the presence of noise. By computer simulations, it was found that the estimation of the model parameters is not sensitive to the parameters' initial estimate (even with greater than 100% error) if the initial estimate of the tissue mixtures is within 10% error from the phantom values. The MRF penalty on the tissue mixture spatial distribution is necessary to ensure the convergence of the iterative tissue mixture estimation in the case of noise level proportional to mean (i.e., similar to Poisson noise). The noise level and initial estimate error are fully within practical conditions, demonstrating that the MAP-EM algorithm is potentially valid in practice. It provides a theoretical or deterministic solution to the PV effect, and its successful implementation could improve quantitative volumetric and feature analyses.
    Full-text · Conference Paper · Dec 2007